Nonlinear Delay Differential Equations and Their Application to Modeling Biological Network Motifs ✉ David S
Total Page：16
File Type：pdf, Size：1020Kb
ARTICLE https://doi.org/10.1038/s41467-021-21700-8 OPEN Nonlinear delay differential equations and their application to modeling biological network motifs ✉ David S. Glass 1, Xiaofan Jin 2 & Ingmar H. Riedel-Kruse 3 Biological regulatory systems, such as cell signaling networks, nervous systems and ecolo- gical webs, consist of complex dynamical interactions among many components. Network motif models focus on small sub-networks to provide quantitative insight into overall 1234567890():,; behavior. However, such models often overlook time delays either inherent to biological processes or associated with multi-step interactions. Here we systematically examine explicit-delay versions of the most common network motifs via delay differential equation (DDE) models, both analytically and numerically. We find many broadly applicable results, including parameter reduction versus canonical ordinary differential equation (ODE) models, analytical relations for converting between ODE and DDE models, criteria for when delays may be ignored, a complete phase space for autoregulation, universal behaviors of feedforward loops, a unified Hill-function logic framework, and conditions for oscillations and chaos. We conclude that explicit-delay modeling simplifies the phenomenology of many biological networks and may aid in discovering new functional motifs. 1 Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot, Israel. 2 Gladstone Institutes, San Francisco, CA, USA. 3 Department of Molecular and Cellular Biology, and (by courtesy) Departments of Applied Mathematics and Biomedical Engineering, University of Arizona, Tucson, AZ, USA. ✉ email: [email protected] NATURE COMMUNICATIONS | (2021) 12:1788 | https://doi.org/10.1038/s41467-021-21700-8 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21700-8 iological regulation consists of complex networks of is a corresponding DDE, where x(t − τ) represents the value of x dynamical interactions1–5. Transcription factors control at a time τ units in the past, making the effect of x on the current B 6,7 production of other transcription factors , kinases the rate of change of x delayed by a time τ. The time scales and delays activation of other kinases1,3,4, cells the growth of other cells8,9, are thus explicit (Fig. 1), better capturing dynamics where and species the population size of other species10. “Network intermediate steps are not fast, as compared to ODE models, motifs” describe particularly important substructures in biological which may fail to capture real delays, hide their effects by over- networks, such as negative feedback, feedforward regulation and simplification, or require additional variables and parameters to cascades2,6,11. The function of network motifs often depends on predict complex phenomena17,23,32. Such delay models have been emergent properties of many parameters, notably delays7,12,13, productive in a variety of biological scenarios, such as processes and previous work showed that delays can lead to dramatic with many intermediate but unimportant steps33,34 and age- biological changes, such as sustained oscillations12,14–17. Delays structured populations34,35. Network models with delays have are crucial to behavior in natural18–20 and synthetic21,22 genetic also been explored in various contexts36–39. Intercellular signal oscillators, in development and disease2,7,14,16,17,23–31, and in transduction, for example, requires multiple time-consuming ecological booms and busts8. Despite such examples, there has steps that are not negligible23,32. been no thorough treatment of network motifs with delays With DDEs, multiple steps (“cascades”) within a network can included explicitly. be rigorously simplified into a single step with delay (see Here we provide a theoretical and practical basis for analyzing below), an approach which has been explored in a variety of network motifs with explicit delays, and we demonstrate the biological contexts18,33,34,40–42. This makes interpretation of utility of this approach in a variety of biological contexts. the phenomenology simpler than with ODEs and reduces the Whereas network motifs are typically modeled using ordinary number of equations and parameters in the model18,23.This differential equations (ODEs) with variables reacting to one idea can be visualized by depicting regulatory networks as another instantaneously (timed by rate constants), we explore directed graphs, with nodes representing biological species, and network motif modeling using delay differential equations pointed vs. blunt arrows indicating activation vs. repression, (DDEs), which have derivatives depending explicitly on the value respectively (Fig. 1). DDE models allow a single arrow to of variables at times in the past. For example, faithfully capture many biochemical steps18,33,43,expandingthe available dynamics in a model with a reduced set of equations x_ðtÞ¼αxðtÞð1Þ and parameters. DDE regulatory models have in fact been used is an ODE, and widely to model a range of biological phenomena such 17,44 30,31 _ð Þ¼α ð À τÞðÞ as development and hematopoiesis . For instance, a 1- x t x t 2 variable ODE such as Eq. (1) can only produce exponential ODE Model DDE Model a τ τ Z XX Y Z X X Z b c 1.0 1.0 0.8 0.8 0.6 0.6 Concentration 0.4 Concentration 0.4 0.2 0.2 x y z 0.0 0.0 0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 Time (t ) Time (t ) Fig. 1 Explicit inclusion of time delays in mathematical models of network motifs can reproduce non-delay models using fewer variables and parameters, but can also lead to more complex behavior. a An example genetic regulatory network including genes X, Y, and Z, regulated by one another either positively (activation, pointed arrow) or negatively (repression, blunt arrow). The model on the left incorporates no explicit delay terms, whereas the model on the right incorporates explicit delay terms τx and τz. b Ordinary differential equations (ODEs) and delay differential equations (DDEs) corresponding to (a) with regulation strengths αi, removal rates βi and cooperativities ni. c Numerical simulations of equations in (b) for one set of initial conditions using parameters αx = 1, αy = 1.2, αz = 1.2, βx = βy = βz = 1, nx = ny = nz = 2 for the ODE simulation, and αx = 0.5, αz = 0.3529, βx = βz = 0.5, nx = nz = 2 for the DDE simulations (dashed curves: τx = 0.8, τz = 0.1; solid curves: τx = 3, τz = 4). Note that delays can cause more complex dynamics (e.g., transient oscillations) compared to models in which effects are instantaneous, and where both models give the same long-term steady state behavior. Left: instantaneous effects modeled using ODEs, Right: delayed effects modeled using DDEs. 2 NATURE COMMUNICATIONS | (2021) 12:1788 | https://doi.org/10.1038/s41467-021-21700-8 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-21700-8 ARTICLE growth or decay45,46. A corresponding DDE, such as Eq. (2), a fi can also oscillate (with amplitude approaching zero or in nity) X Y and, if nonlinear, lead to stable oscillations, bistability and chaos13,26,47–49. A key challenge in using DDEs is their mathematical com- b plexity relative to ODEs50,51. For example, while Eq. (1)is1- dimensional because one initial condition (x at t = 0) determines X Y X Y all future behavior, Eq. (2)isinfinite-dimensional, because x must be specified at all times − τ ≤ t ≤ 0 to predict a unique c 6 6 solution50,51. Concretely, the solution to Eq. (1) can be written as a single exponential, while Eq. (2) generally must be specified as a 5 5 sum of an infinite number of exponential terms to satisfy initial 4 4 conditions. 3 3 Despite the challenges, much progress has been made in 46,52,53 2 2 Concentration analytical understanding of DDEs , and numerical Concentration 54,55 methods exist for simulation . We thus see an opportunity to 1 1 use DDEs to recapitulate dynamics found in ODE solutions of X Y network motif behavior with fewer genes and thus fewer mod- 0 0 eling parameters and equations (Fig. 1), a type of “modeling 0 2 4 6 8 10 0 2 4 6 8 10 Time (T) Time (T) simplicity.” In this work, we thoroughly examine the most common net- Fig. 2 Direct regulation by activators and inhibitors can be captured work motifs2,6 with explicit delays and present an approachable, using a unified delay model. a The simplest regulation network consists of step-by-step view of the mathematical analysis (applying estab- a single input X directly regulating a single output Y. We use a dotted arrow lished DDE methods34,51) in order to make such delay equations to represent either activation or repression, with an implied explicit delay. easy to use for biologists and others. We show the essential role of b Direct activation (left) is represented by a pointed arrow and direct delays in autoregulation, feedforward loops, feedback loops, repression (right) by a blunt arrow. c Time response (Eq. (7)) of activated multiple feedback, and complex networks, as well as instances (left) and repressed (right) Y following a rise in X that exponentially where delays may be ignored. A reference summary is provided at approaches a new steady state (ηX = 6) from zero (governed by _ the end in Table 2. Finally, we discuss how these results can be XðTÞ¼ηX À XðTÞ). Note that the finite value of ηX leads to an effective applied to understanding fundamental design principles of var- leakage slightly greater than ϵ for the repressor case. For the activator ious natural biological systems. (n = −2), we chose η = 5.6528 to match esthetically between the X and n Y steady states, such that ηX ¼ ϵ þ η=ð1 þ ηXÞ. For the repressor (n = 2), we similarly chose η = 5.5 to match esthetically between the X steady state Results and Y initial condition, such that ηX = η + ϵ.